Re: [PATCH 2/2] x86: Shorten RESET_CALL_DEPTH
From: Peter Zijlstra
Date: Mon May 15 2023 - 06:26:30 EST
On Mon, May 15, 2023 at 10:47:42AM +0100, Andrew.Cooper3@xxxxxxxxxx wrote:
> On 15/05/2023 10:28 am, Peter Zijlstra wrote:
> > Shrink it by 4 bytes:
> >
> > 0: 31 c0 xor %eax,%eax
> > 2: 48 0f ba e8 3f bts $0x3f,%rax
> > 7: 65 48 89 04 25 00 00 00 00 mov %rax,%gs:0x0
> > #define RESET_CALL_DEPTH \
> > - mov $0x80, %rax; \
> > - shl $56, %rax; \
> > + xor %eax, %eax; \
> > + bts $59, %rax; \
>
> $63 ?
>
> The disassembly looks correct.
Yeah, uhmm, clearly I fixed it somewhere but not on the version I send
out :-(
Too bad we need the RAX.W prefix...
---
Subject: x86: Shorten RESET_CALL_DEPTH
From: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Date: Fri Feb 10 10:10:57 CET 2023
RESET_CALL_DEPTH is a pretty fat monster and blows up UNTRAIN_RET to
20 bytes:
19: 48 c7 c0 80 00 00 00 mov $0x80,%rax
20: 48 c1 e0 38 shl $0x38,%rax
24: 65 48 89 04 25 00 00 00 00 mov %rax,%gs:0x0 29: R_X86_64_32S pcpu_hot+0x10
Shrink it by 4 bytes:
0: 31 c0 xor %eax,%eax
2: 48 0f ba e8 3f bts $0x3f,%rax
7: 65 48 89 04 25 00 00 00 00 mov %rax,%gs:0x0
Shrink RESET_CALL_DEPTH_FROM_CALL by 5 bytes by only setting al, the
other bits are shifted out (the same could be done for
RESET_CALL_DEPTH, but the xor+bts sequence has less depencies due to
the zeroing).
Suggested-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
---
arch/x86/include/asm/nospec-branch.h | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
--- a/arch/x86/include/asm/nospec-branch.h
+++ b/arch/x86/include/asm/nospec-branch.h
@@ -84,12 +84,12 @@
movq $-1, PER_CPU_VAR(pcpu_hot + X86_call_depth);
#define RESET_CALL_DEPTH \
- mov $0x80, %rax; \
- shl $56, %rax; \
+ xor %eax, %eax; \
+ bts $63, %rax; \
movq %rax, PER_CPU_VAR(pcpu_hot + X86_call_depth);
#define RESET_CALL_DEPTH_FROM_CALL \
- mov $0xfc, %rax; \
+ movb $0xfc, %al; \
shl $56, %rax; \
movq %rax, PER_CPU_VAR(pcpu_hot + X86_call_depth); \
CALL_THUNKS_DEBUG_INC_CALLS